26.7 Convergence in Measure

Definition 26.7.1 (In Measure).label Let $(X, \cm, \mu)$ be a measure space and $(Y, d)$ be a separable metric space. For each $\eps, \delta > 0$, let

\[U(\delta, \eps) = \bracs{(f, g) \in \mathcal{L}^0(X; Y)| \mu\bracs{d(f, g) > \delta} < \eps}\]

then

\[\fB = \bracs{U(\delta, \eps)|\eps, \delta > 0}\]

forms a fundamental system of entourages for a uniformity. The uniformity induced by $\fB$ is the uniform structure of convergence in measure on $\mathcal{L}^{0}(X; Y)$.

Proof. It is sufficient to check the conditions of Proposition 6.1.8:

  1. (FB1)

    For each $\eps, \eps', \delta, \delta' > 0$,

    \[U(\delta \wedge \delta', \eps \wedge \eps') \subset U(\delta, \eps) \cap U(\delta', \eps')\]

  2. (UB3)

    For each $\eps, \delta > 0$ and $f, g, h \in \mathcal{L}^{0}(X; Y)$,

    \[\bracs{d(f, h) > \delta}\subset \bracs{d(f, g) > \delta}\cup \bracs{d(g, h) > \delta}\]

    so $U(\delta/2, \eps/2) \circ U(\delta/2, \eps/2) \subset U(\delta, \eps)$.

$\square$

Definition 26.7.2 (Ky Fan Metric).label Let $(X, \cm, \mu)$ be a measure space, $(Y, d)$ be a separable metric space, and

\[\alpha: L^{0}(X; Y)^{2} \to [0, \infty) \quad (f, g) \mapsto \inf\bracs{\eps > 0| \mu\bracs{d(f, g) > \eps} \le \eps}\wedge 1\]

then:

  1. (1)

    $\alpha$ is a metric on $L^{0}(X; Y)$.

  2. (2)

    $\alpha$ induces the uniform structure of convergence in measure on $L^{0}(X; Y)$.

The mapping $\alpha$ is the Ky Fan metric on $L^{0}(X; Y)$.

Proof. (1): Let $f, g, h \in \mathcal{L}^{0}(X; Y)$, then

  1. (M)

    If $\alpha(f, g) = 0$, then by continuity from above,

    \[\mu\bracs{d(f, g) > 0}= \limv{n}\mu\bracs{d(f, g) > 1/n}\le \limv{n}\frac{1}{n}= 0\]

    so $f = g$ almost everywhere.

  2. (PM3)

    For each $\eps > 0$,

    \[\bracs{d(f, h) > \eps}\subset \bracs{d(f, g) > \eps/2}\cup \bracs{d(g, h) > \eps/2}\]

    so $\alpha(f, h) \le \alpha(f, g) + \alpha(g, h)$.

so $\alpha$ is a metric on $\mathcal{L}^{0}(X; Y)$, modulo almost everywhere equality.

(2): Let $f, g \in \mathcal{L}^{0}(X; Y)$. For any $\eps, \delta > 0$, if $\alpha(f, g) < \eps \wedge \delta$, then there exists $r \in (0, \eps \wedge \delta]$ such that $\mu\bracs{d(f, g) > r}\le r$. Thus

\[\bracs{(f, g) \in \mathcal{L}^0(X; Y)|\alpha(f, g) < \eps \wedge \delta}\subset \bracs{(f, g) \in \mathcal{L}^0(X; Y)|\mu\bracs{d(f, g) >\delta} < \eps}\]

On the other hand, if $\mu\bracs{d(f, g) > \eps}\le \eps$, then $d(f, g) \le \eps$. Therefore $\alpha$ induces the uniform structure of convergence in measure.$\square$

Lemma 26.7.3.label Let $(X, \cm, \mu)$ be a finite measure space, $(Y, d)$ be a separable metric space, and $\seq{f_n}$ and $f$ be Borel measurable functions from $X$ to $Y$ such that $f_{n} \to f$ almost everywhere, then $f_{n} \to f$ in measure.

Proof. Let $\eps > 0$, then for almost every $x \in X$, there exists $N \in \natp$ such that $d(f_{n}(x), f(x)) < \eps$ for all $n \ge N$, so

\[\mu\paren{\bigcup_{N \in \natp}\bigcap_{n \ge N}\bracs{d(f_n, f) < \eps}}= \mu(X)\]

By continuity from above (Proposition 23.1.5),

\[\limv{N}\mu{\bracs{d(f_N, f) \ge \eps}}\le \limv{N}\mu\paren{\bigcup_{n \ge N}\bracs{d(f_n, f) \ge \eps}}= 0\]

$\square$

Theorem 26.7.4.label Let $(X, \cm, \mu)$ be a measure space and $(Y, d)$ be a Polish space, then:

  1. (1)

    For any $\seq{f_n}\subset L^{0}(X; Y)$ that is Cauchy in measure, there exists $f \in L^{0}(X; Y)$ and a subsequence $\seq{n_k}$ such that $f_{n_k}\to f$ almost everywhere and in measure.

  2. (2)

    $L^{0}(X; Y)$ equipped with the uniform structure of convergence in measure is complete.

Proof, [Theorem 2.30, Fol99]. (1): Since $\seq{f_n}$ is Cauchy in measure, there exists a subsequence $\seq{n_k}$ such that for each $k \in \natp$, $\mu(\bracsn{d(f_{n_k}, f_{n_{k+1}}) > 2^{-k}}) \le 2^{-k}$.

In this case, for any $K \in \natp$ and $j \ge k \ge K$,

\[\bracsn{d(f_{n_j}, f_{n_k}) > 2^{-K+1}}\subset \bigcup_{\ell = k}^{j - 1}\bracsn{d(f_{n_\ell}, f_{n_{\ell + 1}}) > 2^{-\ell}}\subset \bigcup_{\ell = K}^{\infty}\bracsn{d(f_{n_\ell}, f_{n_{\ell + 1}}) > 2^{-\ell}}\]

By monotonicity and subadditivity,

\[\mu\paren{\bigcup_{j, k \ge K}\bracsn{d(f_{n_j}, f_{n_k}) > 2^{-K+1}}}\le \sum_{\ell \ge K}\mu\bracsn{d(f_{n_\ell}, f_{n_{\ell + 1}}) > 2^{-\ell}}\le 2^{-K+1}\]

so

\[\mu\braks{\paren{\bigcap_{j, k \ge K}\bracsn{d(f_{n_j}, f_{n_k}) \le 2^{-K+1}}}^c}\le 2^{-K+1}\]

By the First Borel-Cantelli Lemma,

\[\mu\braks{\limsup_{K \to \infty}\paren{\bigcap_{j, k \ge K}\bracsn{d(f_{n_j}, f_{n_k}) \le 2^{-K+1}}}^c}= 0\]

Thus, for almost every $x \in X$, there exists $K \in \natp$ such that $d(f_{n_j}(x), f_{n_k}(x)) < 2^{-K+1}$ for all $j, k \ge K$. Therefore $\seq{f_n(x)}$ is Cauchy for almost every $x$, and converges almost everywhere to a Borel measurable function $f \in L^{0}(X; Y)$.

Finally, for each $K \in \natp$,

\[\mu\bracs{d(f_{n_K}, f) > 2^{-K+1}}\le \sum_{k \ge K}\mu\bracs{d(f_{n_k}, f_{n_K}) > 2^{-k}}\le \sum_{k \ge K}2^{-k}\]

so $f_{n_k}\to f$ in measure as well.

(2): Since the uniform structure of convergence in measure on $L^{0}(X; Y)$ is defined by the Ky Fan metric, completeness follows from (1) and Proposition 8.1.4.$\square$

Post a Comment

Name:Email:
Please enter the tag of the current page (J5) to post the comment.
Tag: